contact us
Research

In Drone Imagery Data, Don’t Confuse Sampling for Measurement

Bakker,Alex_jpg
by Alex Bakker

At the DroneDeploy conference in San Francisco this year, I had several interesting discussions with members of construction, mining and agriculture companies that are using drones in day-to-day operations. These industries are among the early adopters of drones for aerial imagery and surveying work. However, while the imagery and 3D data generated on their drones may appear to be similar, the way the data are being used means the interpretations—and relevant data business value—can be quite different.

In site-survey data used in construction and mining, ground control points (GCPs), Real Time Kinematic GPS, and LiDAR are used to generate highly accurate 3D models and measurements. In many cases, models built this way can be accurate up to less than six inches, and GPS resolutions often come in even better, at the half-inch accuracy level. This degree of accuracy means the results can be integrated into building systems and CAD models and can be used in place of traditional surveying methods.

In agriculture, use of normalized difference vegetation index (NDVI), multi-spectral cameras, infrared or near infrared (NIR) and other crop imaging technologies may have similar degrees of precision, but because crop growth is inherently a time-series phenomenon and the drone imagery is designed to aid in measuring change over time, this form of imagery is called sampling, not measurement.

As with any sampling protocol, normalizing data is important. This means ensuring variables are controlled or normalized across multiple samples. In the case of agricultural imaging, this can mean controlling for temperature, rainfall, fertilizer, sunlight, time of day, white balance of the imagery, humidity, wind, and other measurements that can affect the imagery. Use of this type of data without accurate controls during collection or normalization will make the measurements less reliable, especially for any modeling or predictive analytics based on these samples.

The major difference here is the intended use of the data—whether it should be compared to itself or other static points (measurements), or whether it is intended to be compared against past and future measurements (sampling). The majority of drone business uses so far involve the collection of aerial imagery to inform decision making. Making sure the necessary information is collected alongside the imagery at the time of capture is critical to building reliable data on which to base those decisions.

About the Author

Alex Bakker conducts research and writes about emerging trends and markets for ISG Insights. He focuses specifically on integration, analytics, social business and super emerging technologies, including the Internet of Things, 3D printing, virtual and augmented reality and drones. Alex manages and analyzes survey data, conducts briefings and interviews and publishes research documents as a part of the ISG Insights team.